Abstract: An ensemble method based on similar weights (also called "two-step method") was used to integrate the quantitative precipitation forecasts of the 12 numerical prediction models, and put them into the operational application. The results indicate that by comparing the TS scores of 24 hours precipitation forecast from May 1 to October 20, 2009, the accuracy of precipitation forecast by the ensemble method is higher than by each single method of 12 ensemble members and by the ensemble average. Further experiments show that the effects of lag time and expanding grid number on the ensemble forecasts are significant, while that of the each single ensemble member is not significant. The application plan is modified according to these experiments: the expanded grid numbers depends on forecast period validity and precipitation level rather than similarity and if using the newest forecast method of ensemble members. The modified plan is better than the original plan, especially for long forecast period validity and high precipitation level. For Example, TS score of 72 hours forecast period validity for 25 mm precipitation level could increase more than 20%.

Key words: Numerical forecast, Precipitation forecast, Ensemble method, Operation application